Potential energy saving in wireline telecoms


Published on

  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Potential energy saving in wireline telecoms

  1. 1. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® ENERGY EFFICIENCY IN COMMUNICATIONS The Potential Impact of Green Technologies in Next-Generation Wireline Networks: Is There Room for Energy Saving Optimization? Raffaele Bolla and Franco Davoli, DIST-University of Genoa Roberto Bruschi, National Inter-University Consortium for Telecommunications Ken Christensen, University of South Florida Flavio Cucchietti, Telecom Italia Suresh Singh, Portland State University ABSTRACT cated architectures able to perform increasingly complex operations in a scalable way. For Recently, network operators around the world instance, high-end routers are increasingly based reported statistics of network energy requirements on complex multirack architectures, which pro- and the related carbon footprint, showing an vide more and more network functionalities; his- alarming and growing trend. Such high energy toric data from manufacturers’ datasheets show consumption can be mainly ascribed to networking continuously raising capacities, by a factor of 2.5 equipment designed to work at maximum capacity every 18 months [1]. However, silicon technolo- with high and almost constant dissipation, inde- gies improve their energy efficiency at a slower pendent of the traffic load. However, recent devel- pace following Dennard’s law (i.e., by a factor of opments of green network technologies suggest 1.65 every 18 months) with respect to routers’ the chance to build future devices capable of capacities and traffic volumes. adapting their performance and energy absorption In the last few years, telcos, ISPs, and public to meet actual workload and operational require- organizations around the world reported statis- ments. In such a scenario, this contribution aims tics of network energy requirements and the at evaluating the potential impact on next-genera- related carbon footprint, showing an alarming tion wireline networks of green technologies in and growing trend. The Global e-Sustainability economic and environmental terms. We based our Initiative (GeSI) [2] estimated an overall net- impact analysis on the real network energy-effi- work energy requirement of about 21.4 TWh in ciency targets of an ongoing European project, 2010 for European telcos, and foresees a figure and applied them to the expected deployment of of 35.8 TWh in 2020 if no green network tech- Telecom Italia infrastructure by 2015–2020. nologies (GNTs) are adopted. The sole introduc- tion of novel low-consumption silicon INTRODUCTION technologies clearly cannot cope with such trends and be sufficient for drawing current network Only recently, telecom operators (telcos) and equipment toward a greener future Internet. Internet service providers (ISPs) have raised their Moreover, it is well known that network links interest in energy efficiency for wired networks and devices are provisioned for busy or rush hour and service infrastructures, making it a high-priori- load, which typically exceeds their average utiliza- ty objective. This interest is motivated by the tion by a wide margin. While this margin is sel- increase in energy prices, the continuing growth of dom reached, nevertheless the overall power the customer population, the spreading of broad- consumption in today’s networks is determined by band access, and the expanding services offered. it and remains more or less constant even in the Indeed, the number of new services being offered presence of fluctuating traffic loads. This situation and the increase in the volume of data traffic fol- suggests the possibility of adapting network ener- low Moore’s law, doubling every 18 months. gy requirements to actual traffic profiles. To support new-generation network infras- Motivated by these considerations, we recent- tructures and related services for a rapidly grow- ly established a partnership with a group of pri- ing customer population, telcos and ISPs need mary device manufacturers and telcos to launch an ever larger number of devices, with sophisti- the ECONET initiative [3], an integrated project 80 0163-6804/11/$25.00 © 2011 IEEE IEEE Communications Magazine • August 2011C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  2. 2. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® (a) The main aim of Power Overall consumption ECONET is to design Number of devices consumption (W) (GWh/year) and develop Home 10 17,500,000 1533 innovative solutions Access 1280 27,344 307 and device Metro/transport 6000 1750 92 prototypes for Core 10,000 175 15 wired network Overall network consumption 1947 infrastructures (from customer-premises (b) equipment to Number of customers per DSLAM 640 backbone switches Home access Usage of a network access (user up time) 30% Link utilization when a user is connected 10% and routers) within 2013. Redundancy degree for metro/transport devices ( t) 13% Redundancy degree for core devices ( c) 100% Core, transport, Redundancy degree of metro/transport links ( t) 100% and metro Redundancy degree of core device links ( c) 50% Link utilization in metro networks 40% Link utilization in core networks 40% (c) Data plane ( d) Control plane ( c) Cooling/power supply ( p) Home 79% 3% 18% Access 84% 3% 13% Metro/transport 73% 13% 14% Core 54% 11% 35% Table 1. a) 2015–2020 network forecast: device density and energy requirements in the business as usual (BAU) case, example based on the Italian network; b) traffic and topological data (average figures) for the 2015–2020 prospective network (source: Telecom Italia); c) internal sources of energy consumption (source: the ECONET Consortium). (IP) funded by the European Commission (EC). are proposing, along with other contributions by The main aim of ECONET is to design and the research community, for disruptively reduc- develop innovative solutions and device proto- ing the network carbon footprint. We report an types for wired network infrastructures (from analysis estimating the impact of GNTs in both customer premises equipment to backbone economical and environmental terms. Conclu- switches and routers) by 2013. The resulting net- sions are then drawn. work platforms will adopt GNTs for aggressively modulating power consumption according to actual workloads and service requirements. A REFERENCE SCENARIO: In this scenario, the goal of the present arti- cle is to evaluate the potential gain to be derived THE ITALIAN CASE from the application of GNTs in quantitative The considered reference scenario is shown in terms. To this aim, we want to assess how tech- Table 1a, which contains the number of devices nological solutions able to trade device perfor- per network segment and their energy consump- mance for power consumption can be effectively tion in the business-as-usual case (i.e., the case used in the short term for reducing the carbon in which no green enhancements are included in footprint and operating expenditures (OPEX) of network devices). Both end-user and operator next-generation wireline networks. In order to equipment have been taken into account. make our impact analysis as realistic as possible, We refer to a network with 17.5 million cus- we consider the energy efficiency targets of the tomers, where we assume the presence of broad- ECONET project, and apply them to a perspec- band-only access technologies, together with tive network of a large-scale telco, which corre- suitable overprovisioning in the metro, transport, sponds to an expected deployment of Telecom and core segments. Italia infrastructure by 2015–2020. Devices’ energy consumption has been forecast The article is organized as follows. We intro- on the basis of present values, high quality specifica- duce a reference scenario based on prospective tions (e.g., European Broadband Code of Conduct), Telecom Italia network development. We briefly and expected “inertial” technological improvements describe some of the most promising GNTs we (e.g., Dennard’s law [1]). In the devices’ energy IEEE Communications Magazine • August 2011 81C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  3. 3. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® requirements, we included the contribution of site behind these algorithms is simple: the upstream The largest part of cooling and powering systems, which account for 36 interface on a link maintains a window of inter- percent of direct device consumption. packet arrival times. This information is then approaches Starting from the scenario in Table 1, the per- used to determine the length of time for which undertaken is user average energy requirement consists of about an interface can be put to sleep such that, with a founded on a few 111 kWh/year, mainly due to home and access high probability, the buffers at that interface will networks, for 79 and 16 percent, respectively. not overflow. The constraint is the non-zero time basic concepts, Metro/transport and core networks account only it takes for the downstream interface to wake which have been for 5 percent, but their joint energy requirement up. The results obtained with real traces show of about 107 GWh/year (about a quarter of the that for loads up to 30 percent of link capacity, generally inspired telco’s direct energy consumption) can be a con- considerable energy savings can be achieved. by energy-saving vincing driver for reducing the carbon footprint of Similarly, the research of Christensen et al. has mechanisms and backbone devices in the near future. specifically addressed how to reduce direct energy Table 1b defines key parameters that allow use of Ethernet links, and has contributed to the power management synthetically representing the average usage of development of the IEEE 802.3az standard. In [7] criteria that are network devices and links. This has been done they first explored the notion of an adaptive link by defining the expected (by 2015–2020) average rate (ALR) for Ethernet, whereby a link would already partially customer up times and loads, the average traffic operate at a low data rate during periods of low adopted in utilization on metro/transport and core net- utilization and at high data rate only for high uti- computing systems. works, and the number of devices and links that lization periods. Given that most links are highly are usually deployed for redundancy purposes. It underutilized, with ALR most Ethernet links is worth noting that the traffic load values in could operate at a low data rate (and thus reduce Table 1b are significantly larger (and conse- energy consumption compared to operation at a quently give rise to conservative consumption high data rate) most of the time [8]. estimations) than those of the current network An implementation of ALR would entail an and indicated in other studies [1]. Ethernet interface having two physical layer Finally, the values in Table 1c, which subdi- implementations and switching between them. vides the equipment consumption into its main The time to switch between physical layer imple- functions/building blocks, outline how the most mentations was deemed to be a major issue, energy-starving elements in network devices resulting in an alternative LPI approach [9] pro- reside in the data plane. In fact, data plane ener- posed by Intel. LPI is the approach specified in gy shares range between 54 and 84 percent, the emerging 802.3az standard, and currently against 13–35 percent for air cooling and power allows a 10 Gb/s link to wake up in less than 3 s. supply for onboard components, and 3–13 percent In [10], Bolla et al. analyzed and empirically for control plane ones. Thus, it is not so surpris- modeled the energy modulation capabilities of ing that the current GNT proposals in this field processing engines in Linux-based software routers are especially devoted to reducing the energy con- equipped with general-purpose and multicore pro- sumption of the network devices’ data plane. cessors that already include LPI and adaptive rate primitives. The results achieved were obtained by evaluating several hardware architectures, and GREEN NETWORK TECHNOLOGIES they suggest that such technologies permit the The largest part of approaches undertaken is trade-off between power consumption and net- founded on a few basic concepts, which have work performance to scale almost linearly. been generally inspired by energy-saving mecha- In [11], the authors extended their approach nisms and power management criteria that are by introducing a control framework for optimally already partially adopted in computing systems. tuning LPI and adaptive rate mechanisms in Following the taxonomy proposed in [1], these order to statistically meet current traffic loads basic concepts can be classified as dynamic power and service requirements. The results obtained scaling and smart standby approaches. on real traffic traces show that energy gains up to 60 percent are feasible. DYNAMIC POWER SCALING Working on similar platforms, Intel Power scaling capabilities allow dynamically reduc- researchers [12] especially focused on LPI primi- ing the working rate of processing engines or link tives and performed a comprehensive study of the interfaces. This is usually accomplished by adopt- impact of transition times on LPI as a function of ing two basic techniques: adaptive rate and Low load. They showed that as the transition times Power Idle (LPI). The former allows dynamically shrink from the value of 10 ms to 1 ms and then modulating the capacity of a link or a processing further to 100 s, the time spent sleeping at 30 engine in order to meet traffic load and service percent load goes from 0 at transition time of 10 requirements. The latter forces links or processing ms to 40 percent when this time is 1 ms, and to 70 engines to enter low-power states when not send- percent when the transition happens in 100 s. ing/processing packets and quickly switch to a high-power state when sending one or more pack- SMART STANDBY ets. These techniques are not exclusive and can be Sleeping and standby approaches are founded on jointly adopted in order to adapt system perfor- power management primitives that allow devices mance to current workload requirements. or parts of them turning themselves almost com- In previous work [4–6], some of the authors pletely off and entering very low energy states, of the present article first faced the energy effi- while all their functionalities are frozen. ciency issue in wireline networks and developed The widespread adoption of this kind of ener- several algorithms that used traffic prediction to gy-aware capability is generally hindered by the put links to low power idle modes. The idea necessity of maintaining the “network presence” 82 IEEE Communications Magazine • August 2011C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  4. 4. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® depending on their current workload and on Standby ( ) 85% their actual usage status. From a simplified point The widespread of view, we characterized such composite energy Performance scaling ( ) 50% profile for a generic network device by consider- adoption of this kind ing the following power breakdown: of energy-aware Network-wide control ( ) 20% • P full is the power absorption of an entirely capability is generally busy device, and corresponds to no green Air cooling/power supply ( ) 15% optimization. hindered by the • Pidle is the power absorption of a device that necessity to maintain Table 2. Green efficiency degrees targeted in the is active, but not performing any opera- ECONET project. tions. the “network pres- • Pstandby is the power absorption of a device ence” of the device. in low-energy standby modes; only basic Network hosts and of the device. Network hosts and devices must operations to maintain the network pres- maintain network connectivity or they will liter- ence are performed. devices must main- ally “fall off the network,” become unreachable, We assume that, as experimentally demon- tain network con- and network applications and services will fail. strated on software router architectures in [10], In [13, 14], Christensen et al. first explored power consumption values between the “idle” nectivity or else they the idea of using a proxy to “cover” for a net- and “full” cases vary linearly with respect to the will literally “fall off work host and thus allow it to go to sleep. Specif- actual workload. the network.” ically, they addressed requirements for a proxy The specific values of Pidle and Pstandby depend to be able to respond to ARP packets on behalf on the efficiency of power scaling and standby of a sleeping host (to maintain reachability from primitives that will be adopted in future network the router), and to respond to other protocol devices. and application messages as needed to maintain In order to provide a solid basis for our esti- full network presence. A specific focus of their mation, we decided to use the same efficiency research was on peer-to-peer (P2P) protocols degrees that are targeted by the ECONET pro- and how they might be proxied to allow for a PC ject. Table 2 reports such minimum efficiency sharing files to sleep most of the time. The prox- targets. Taking the full consumption of the ying requirements they developed led to the cre- device as a term of comparison, the and ation of the European Computer Manufacturers parameters represent how much energy can be Association (ECMA) TC38-TG4 standard. saved by the data plane hardware in standby and Apple has recently announced a product with idle modes, respectively. The parameter weighs green proxying capabilities. an indirect gain on the device’s power supply Nonetheless, the development of such net- and air cooling due to the use of power scaling work-specific low-energy modes is a fundamental primitives. Finally, is meant to represent the key factor for reducing the carbon footprint inside impact of network-wide control strategies (e.g., networks as well, since it will allow switching traffic engineering) for optimizing the energy- some links, entire network devices, or parts there- aware configuration of the overall network at of to a sleep mode in a smart and effective way. transport and core levels. This is the main idea behind emerging approach- Starting from these considerations and the es to network control, routing, and traffic engi- above definitions, we expressed Pidle and Pstandby neering [15], which aim at dynamically putting as follows: network portions to sleep during light utilization periods, in order to minimize the energy require- Pidle = Pfull[(1 – ) d + c + (1 – ) p] (1) ments of the overall network while meeting the operational constraints and current workloads. Pstandby = Pfull[(1 – ) d + c + (1 – ) p] (2) where d, c, and p are defined in Table 1c. IMPACT ANALYSIS Focusing on Eqs. 1 and 2, we can outline how This section is organized as follows. We estimate the power requirements of the data plane in idle the impact of GNTs on energy consumption of and standby states are scaled down by and , future network devices. We try to assess how respectively, while those of the control plane ele- much these technologies can really be exploited. ments are maintained constant and equal to the full Finally, we aggregate all the estimated data for load condition. This is due to the fact that energy- giving a complete overview of the potential aware network devices are meant to keep their pres- impact of GNTs in terms of energy savings, and ence in the network by sending, elaborating, and reduction of both CO2 emissions and OPEX. receiving signaling and control protocol packets. Figure 1 shows the estimated energy profiles THE DEVICE ENERGY PROFILES for energy-aware future devices following the Energy-aware devices are meant to dynamically 2015–2020 forecast in Table 1a. save energy by applying the adaptive capacities of GNTs introduced earlier. POWER SCALING AND STANDBY EXPLOITATION Due to the heterogeneity of functional and In order to estimate the real energy-savings performance requirements of the various kinds coming from a massive adoption of GNTs, we of networking equipment, different exploitations need to know the device energy profiles, and we of green optimizations and technologies can be have also to estimate which and how much stand- reasonably expected. Anyway, the joint adoption by and power scaling primitives would be exploit- of these green optimizations will lead future net- ed in real operating scenarios for both work devices to have a composite energy profile, access/home networks and transport/core ones. IEEE Communications Magazine • August 2011 83C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  5. 5. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® Power consumption (Wh) 15 1500 Power consumption (Wh) 10 1000 5 500 0 0 Full load Idle Standby Full load Idle Standby (a) (b) 8000 15,000 Power consumption (Wh) Power consumption (Wh) 6000 10,000 4000 5000 2000 0 0 Full load Idle Standby Full load Idle Standby (c) (d) Figure 1. a) Energy profiles of next generation green home gateways; b), digital subscriber line access multiplexers (DSLAMs); c) metro/transport; d) core devices. The energy profiles have been obtained by using Eqs. 1 and 2 and the input data in Tables 1 and 2. Focusing on the standby capabilities, it is rea- a core network correspond to about 37.5 percent1 sonable to expect a different exploitation way in of the deployed (and powered) ones. Similar the home and access devices with respect to the remarks can be made also for transport devices, transport/core ones. where the redundancy degree is however usually Regarding end-users and network access, less than the one at the core level. Still according standby modes can be thought to suitably follow to Table 1, redundant transport/metro nodes are the end-user behavior — for example, by tem- about 13 percent of the deployed ones, while each porarily putting home gateways or digital sub- active device usually has a redundant copy for scriber line (DSL) links in standby states when not each of its operating links. In such case, the per- directly used or not needed. Thus, the average centage of operating hardware results being about time, in which home devices, or links in an access 43.5 percent of the network deployment. More- device can be suitably put in standby modes, main- over, green traffic engineering and routing mech- ly depends on end-users’ up-times. Recalling the anisms can further reduce the number of active average end-user up-time (i.e., 30 percent, which links and devices [15], while meeting desired ser- corresponds to about 7 h/day) from Table 2, we vice requirements and traffic loads. Recent stud- can deduce that home gateways and access links ies [15] demonstrate that such policies can lead, can be put in standby for 70 percent of the time. on average, to put more than 20 percent of the Regarding backbone network infrastructures, active devices to sleep. As shown in Fig. 2, by we have to take tighter performance and opera- applying this last “energy gain” we obtained the tional constraints into account. Core, transport average shares of time with reference to the and metro network devices have usually to work access and home networks, and the shares of on large volumes of traffic, and to guarantee top devices and parts of them with reference to the performance in terms of quality of service (QoS), transport and core networks where standby primi- network reliability, and so on. tives can be exploited fruitfully. Here, while real operating devices and parts Regarding the exploitation of power scaling of them cannot be put in standby modes, there is mechanisms, they may be applied to active devices, a large set of redundant devices and links that is and allow modulating power consumption with left powered on only to fast recover from faults. respect to the processed traffic loads. Starting Our main proposal is that devices working at from the discussions earlier, we can easily deduce these levels should include specific standby sup- that the average power need of a generic network port to allow redundant hardware (i.e., entire device using power scaling primitives can be esti- devices, line cards, etc.) to be woken up upon mated to be equal to (1 – )Pidle + Pfull, where fault detection with very fast recovery times. For is the average link (and device) utilization report- these reasons, such elements cannot be simply ed in Table 1b for each network segment. shut off, but they need to maintain their network control and signaling activities to always have THE OVERALL IMPACT up-to-date network data and information. This impact analysis has been performed by con- According to Table 1, at the core network sidering both the devices working inside the 1This value has been level, each active device is coupled with a redun- telco network and the customer premises equip- obtained as (1 – c/2) dant copy, and furthermore each active device has ment (thus, our estimation covers the whole c/2 with reference to a quarter of its network links still in redundancy. wireline network, from home gateway to core Table 1b. Thus, the actually operating hardware elements in routers). For all such devices, we used the ener- 84 IEEE Communications Magazine • August 2011C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  6. 6. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® Standby Standby Standby Active 62.5% Network-wide Active 70% Active 70% Network-wide 30% 37.5% control 30% control (at 10% (at 40% 9.4% (at 40% 7.5% of the of the of the max max max capacity) capacity) capacity) Redundant HW Redundant HW 62.5% 53.1% (a) (b) (c) Figure 2. a) Standby exploitation shares for access and home devices; b) for metro and transport ones; c) for core routers. In case a), such times correspond to the user activity profiles (Table 1b). In the b) and c) cases, standby times arise from both the share of redundant hardware and the devices (and/or parts of them) put in standby modes by green traffic engineering and routing mechanisms. The figures also report the average utilization of devices when active. Full load power Number of Overall full consumption Energy gains Percentage gains consumption (Wh) devices (GWh/year) (GWh/year) Home 10 17,500,000 1533 70% 1060 Access 1,280 27,344 307 70% 213 Metro/transport 6,000 1,750 92 54% 49 Core 10,000 175 15 58% 9 Overall gain 68% Total BAU (GWh/year) 1947 Total gains with green technologies (GWh/year) 1331 Table 3. Impact of green technologies on the 2015–2020 perspective network in terms of energy savings. gy profiles in Fig. 1 by weighing them with the This considerable OPEX reduction especially exploitation shares in Fig. 2. arises from devices working at the access level, Table 3 reports the overall impact of GNTs which, thanks to their numerousness, account for in terms of energy consumption reduction with 78.6 percent of the overall Telco’s gain. Also respect to the BAU scenario, which was intro- GNTs applied to metro/transport and core net- duced in Table 1a. The obtained values show work devices, which account for 18.1 percent and that the energy requirements of the reference 3.3 percent, respectively, provide a non negligible network can be sensibly scaled down by about impact in terms of both OPEX saving and carbon 1331 GWh/year (which roughly equals 956 footprint reduction, especially when compared to ktons/year of CO2 emissions), corresponding to their number in the network. 68 percent of BAU energy consumption — a sig- Finally, Fig. 3 shows how the overall energy nificant figure, comparable (and additional!) to gain varies according to different values of effi- the improvement obtainable by the sole increase ciency degrees for standby and power scaling in hardware efficiency (as predicted by Den- technologies. These results outline that standby nard’s law) with respect to the capacity increase primitives have a sensibly higher impact on the of network devices. final gain than the power scaling ones. This gain especially arises from the customer side, where, by considering only the savings at the home gateways, we obtained a potential reduction CONCLUSION equal to 1060 GWh/year, which corresponds to We propose a forecast of the potential impact of about 70 percent with respect to the BAU GNTs, if massively adopted in a large-scale oper- requirements. The energy gain would be much ator network by 2015–2020. Our estimates consid- larger if we also considered the potential addi- er both the equipment inside the telco network tional savings of GNTs applied to other customer and the customer premises equipment, and we devices like set-top-boxes, VoIP phones, and PCs. target the same short-term energy efficiency goals Also, the telco side energy savings, even as the ECONET European project. The figures though corresponding to about a quarter of those resulting from our calculations on this basis out- at the customer side , show an almost surprising line that GNTs can allow saving about 68 percent impact: a total gain equal to 271 GWh/year (a of energy requirements of the overall network reduction of more than 65 percent with respect and an OPEX reduction for the single reference to the BAU scenario). Using the EIA2 forecast- telco of about $33 million/year, due to a gain of ing on 2020 energy cost, this energy saving can be 271 GWh/year in the energy consumption of wire- 2EIA, U.S. Energy Infor- directly translated to a total OPEX reduction of line infrastructures. These figures are the upper mation Administration, nearly $33 million/year for the reference Telco. bounds of what could be attained, whereas the http://www.eia.doe.gov/. IEEE Communications Magazine • August 2011 85C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®
  7. 7. C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND® aging Energy Consumption Costs in Desktop PCs and LAN Switches with Proxying, Split TCP Connections, and Scaling of Link Speed,” Int’l. J. Network Mgmt., vol. 15, 1600 no. 5, Sept. 2005, pp. 297–310. 1400-1600 1200-1400 [8] C. Gunaratne et al., “Reducing the Energy Consumption of 1000-1200 Ethernet with an Adaptive Link Rate (ALR),” IEEE Trans. 800-1000 1400 Computers, vol. 57, no. 4, Apr. 2008, pp. 448–61. 600-800 [9] R. Hays, “Active/Idle Toggling with 0BASE-x for Energy 400-600 Efficient Ethernet,” presentation to the IEEE 802.3az 200-400 1200 Task Force, Nov. 2007, http://www.ieee802.org/3/az/ 0-200 public/nov07/hays_1_1107.pdf ________________ [10] R. Bolla, R. Bruschi, and A. Ranieri, “Green Support for PC- 1000 based Software Router: Performance Evaluation and Mod- Energy gain (GWh/year) eling,” Proc. IEEE ICC ’09, Dresden, Germany, June 2009. [11] R. Bolla, R. Bruschi, and F. Davoli, “Energy-Aware Per- 800 formance Optimization for Next Generation Green Net- work Equipment,” Proc. ACM SIGCOMM PRESTO ’09, 600 Barcelona, Spain, Aug. 2009. [12] S. Nedevschi et al., “Reducing Network Energy Con- sumption via Sleeping and Rate-Adaptation,” Proc. 400 USENIX NSDI ’08, San Francisco, CA, USA, Apr. 2008. [13] K. Christensen and F. Gulledge, “Enabling Power Man- agement for Network-Attached Computers,” Int’l. J. 200 Network Mgmt., vol. 8, no. 2, Apr. 1998, pp. 120–30. [14] M. Jimeno, K. Christensen, and B. Nordman, “A Network 90 80 Connection Proxy to Enable Hosts to Sleep and Save Ener- 70 0 60 gy,” Proc. IEEE IPCCC ’09, Dec. 2009, pp. 101–10. 50 40 70 [15] J. Restrepo, C. Gruber, and C. Machoca, “Energy Pro- 30 file Aware Routing,” Proc. IEEE GreenComm ’09, Dres- 20 10 35 Standby efficiency 0 den, Germany, June 2009. (%) Power scaling efficiency (%) BIOGRAPHIES Figure 3. Overall energy gain for the whole network (including both customer R AFFAELE B OLLA [M] (raffaele.bolla@unige.it) received his _____________ and telco sides) with respect to different efficiency degrees of standby and Laurea degree in electronic engineering in 1989 and his power scaling mechanisms. Ph.D. degree in telecommunications in 1994, both from the University of Genoa, Italy. He is currently an associate professor at the Department of Communications, Comput- er and Systems Science (DIST) of the University of Genoa. His main current research interests are in the future Inter- actual savings will depend on the extent of the net and green networking. He is the Principal Investigator adoption of the new methodologies. The technol- of the ECONET project. ogy to achieve these savings is there. At this ROBERTO BRUSCHI [M] (roberto.bruschi@cnit.it) received his _____________ point, the only question that remains open is M.Sc. degree in telecommunication engineering in 2002 whether these numbers will be impressive enough and his Ph.D. degree in electronic engineering in 2006 to convince research and industrial communities from the University of Genoa, Italy. He is currently a to set forth the actual development of a greener researcher at the National Inter-University Consortium for Telecommunications (CNIT) at the University of Genoa Internet. Whereas telcos should be encouraged by Research Unit. His main research interests include the the potential OPEX reduction, customers and future Internet, green networking, and software routers. regulators would also support the process once provided with a simple and clear explanation of KEN CHRISTENSEN [SM] (christen@csee.usf.edu) is a professor ____________ in the Department of Computer Science and Engineering at their own cost savings, especially if guaranteed by the University of South Florida. He received his Ph.D. in certification authorities (e.g., Energy Star3). electrical and computer engineering from North Carolina State University in 1991. His primary research interest is in ACKNOWLEDGMENTS green networks. He is a licensed Professional Engineer in the state of Florida, and a member of ACM and ASEE. This work was supported by the European FP7 ECONET project under grant agreement no. FLAVIO CUCCHIETTI (flavio.cucchietti@telecomitalia.it) is with __________________ 258454. Telecom Italia R&D Center in Turin, Italy. He is managing innovative projects on infrastructural and powering solu- REFERENCES tions for the next-generation network. He is currently a member of many standardization groups acting on energy [1] R. Bolla et al., “Energy Efficiency in the Future Internet: A efficiency for telecommunications and IT, and network Survey of Existing Approaches and Trends in Energy-Aware infrastructure. He co-chairs the Energy Efficiency Inter Fixed Network Infrastructures,” IEEE Commun. Surveys and Operators Collaboration Group (GeSi-EE IOCG). He is a Tutorials, vol. 13, no. 2, 2nd qtr. 2011, pp. 233–44. board member of the Italian standardization institute in [2] Global e-Sustainibility Initiative (GeSI), “SMART 2020: the electrical, electronic and telecommunication fields (CEI). Enabling the Low Carbon Economy in the Information Age,” http://www.theclimategroup.org/assets/resources/ FRANCO DAVOLI [SM] (franco.davoli@unige.it) received his Laurea ___________ publications/Smart2020Report.pdf. __________________ degree in electronic engineering in 1975 from the University of [3] The “low Energy COnsumption NETworks” (ECONET) Genoa, Italy. He is currently a full professor of telecommunica- Project, IP project funded by the EC in ICT-Call 5 of the tion networks at DIST, University of Genoa. His current 7th Framework Programme, grant agreement no. research interests are in multiservice networks, green network- 258454, http://www.econet-project.eu. ing, wireless networks, and multimedia communications and [4] M. Gupta and S. Singh, “Greening of the Internet,” services. He is a member of the CNIT Scientific Board. Proc. ACM SIGCOMM ’03, Karlsruhe, Germany, Aug. 2003, pp. 19–26. SURESH SINGH [M] (singh@cs.pdx.edu) received his Ph.D. in com- _________ [5] M. Gupta and S. Singh, “Dynamic Ethernet Link Shut- puter science from the University of Massachusetts at Amherst in down for Power Conservation on Ethernet Links,” Proc. 1990 and his B.Tech., also in computer science, from the Indian IEEE ICC ’07, Glasgow, Scotland, June 2007. Institute of Technology in 1984. He is currently a professor in the [6] M. Gupta, S. Grover, and S. Singh, “A Feasibility Study Department of Computer Science at Portland State University. for Power Management in LAN Switches,” Proc. IEEE His research interests include green technologies for Internet ICNP ’04, Berlin, Germany, Oct. 2004. devices, energy-efficient computation, wireless networking pro- 3 http://www.energystar.gov/ [7] C. Gunaratne, K. Christensen, and B. Nordman, “Man- tocols, information theory, and performance analysis. 86 IEEE Communications Magazine • August 2011C qM IEEE M ommunications q qM Previous Page | Contents | Zoom in | Zoom out | Front Cover | Search Issue | Next Page MqM q Qmags THE WORLD’S NEWSSTAND®